京都大学 / 新制・課程博士 / 博士(情報学) / 甲第25427号 / 情博第865号 / 京都大学大学院情報学研究科社会情報学専攻 / (主査)教授 伊藤 孝行, 教授 黒田 知宏, 教授 岡部 寿男, 吉川 正俊(京都大学 名誉教授) / 学位規則第4条第1項該当 / Doctor of Informatics / Kyoto University / DFAM

In recent years, trusted execution environments (TEEs) have seen increasing deployment in computing devices to protect security-critical software from run-time attacks and provide isolation from an untrustworthy operating system (OS). A trusted party verifies the software that runs in a TEE using remote attestation procedures. However, the publication of transient execution attacks such as Spectre and Meltdown revealed fundamental weaknesses in many TEE architectures, including Intel Software Guard Exentsions (SGX) and Arm TrustZone. These attacks can extract cryptographic secrets, thereby compromising the integrity of the remote attestation procedure. In this work, we design and develop a TEE architecture that provides remote attestation integrity protection even when confidentiality of the TEE is compromised. We use the formally verified seL4 microkernel to build the TEE, which ensures strong isolation and integrity. We offload cryptographic operations to a secure co-processor that does not share any vulnerable microarchitectural hardware units with the main processor, to protect against transient execution attacks. Our design guarantees integrity of the remote attestation procedure. It can be extended to leverage co-processors from Google and Apple, for wide-scale deployment on mobile devices. / Under de senaste åren används betrodda exekveringsmiljöer (TEE) allt mera i datorutrustning för att skydda säkerhetskritisk programvara från attacker och för att isolera dem från ett opålitligt operativsystem. En betrodd part verifierar programvaran som körs i en TEE med hjälp av fjärrattestering. Nyliga mikroarkitekturella anfall, t.ex. Spectre och Meltdown, har dock visat grundläggande svagheter i många TEE-arkitekturer, inklusive Intel SGX och Arm TrustZone. Dessa attacker kan avslöja kryptografiska hemligheter och därmed äventyra integriteten av fjärrattestning. I det här arbetet utvecklar vi en arkitektur för en betrodd exekveringsmiljö (TEE) som ger integritetsskydd genom fjärrattestering även när TEE:s konfidentialitet äventyras. Vi använder den formellt verifierade seL4-mikrokärnan för att bygga TEE:n som garanterar stark isolering och integritet. För att skydda kryptografiska operationer, overför vi dem till en säker samprocessor som inte delar någon sårbar mikroarkitektur med huvudprocessorn. Vår arktektur garanterar fjärrattesteringens integritet och kan utnyttja medprocessorer från Google och Apple för att användas i stor skala på mobila enheter.

In modern computing environments, hardware resources are commonly shared, and parallel computation is more widely used. Users run their services in parallel on the same hardware and process information with different confidentiality levels every day. Running parallel tasks can cause privacy and security problems if proper isolation is not enforced. Computers need to rely on a trusted root to protect the data from malicious entities. Intel proposed the Software Guard eXtension (SGX) to create a trusted execution environment (TEE) within the processor. SGX allows developers to benefit from the hardware level isolation. SGX relies only on the hardware, and claims runtime protection even if the OS and other software components are malicious. However, SGX disregards any kind of side-channel attacks. Researchers have demonstrated that microarchitectural sidechannels are very effective in thwarting the hardware provided isolation. In scenarios that involve SGX as part of their defense mechanism, system adversaries become important threats, and they are capable of initiating these attacks. This work introduces a new and more powerful cache side-channel attack that provides system adversaries a high resolution channel. The developed attack is able to virtually track all memory accesses of SGX execution with temporal precision. As a proof of concept, we demonstrate our attack to recover cryptographic AES keys from the commonly used implementations including those that were believed to be resistant in previous attack scenarios. Our results show that SGX cannot protect critical data sensitive computations, and efficient AES key recovery is possible in a practical environment. In contrast to previous attacks which require hundreds of measurements, this is the first cache side-channel attack on a real system that can recover AES keys with a minimal number of measurements. We can successfully recover the AES key from T-Table based implementations in a known plaintext and ciphertext scenario with an average of 15 and 7 samples respectively.

In recent years, trusted execution environments like Intel SGX have allowed developers to protect sensitive code inside so called enclaves. These enclaves protect its code and data even in the cases of a compromised OS. However, SGX enclaves have been shown to be vulnerable to numerous side-channel attacks. Therefore, there is a need to investigate ways that such attacks against enclaves can be detected. This thesis investigates the viability of using performance counters to detect an SGX-targeting side-channel attack, specifically the recent Load Value Injection (LVI) class of attacks. A case study is thus presented where performance counters and a threshold-based detection method is used to detect variants of the LVI attack. The results show that certain attack variants could be reliably detected using this approach without false positives for a range of benign applications. The results also demonstrate reasonable levels of speed and overhead for the detection tool. Some of the practical limitations of using performance counters, particularly in an SGX-context, are also brought up and discussed.

<div>Driven by breakthroughs in mobile and IoT devices, on-device computation becomes promising. Meanwhile, there is a growing concern over its security: it faces many threats</div><div>in the wild, while not supervised by security experts; the computation is highly likely to touch users’ privacy-sensitive information. Towards trustworthy on-device computation, we present novel system designs focusing on two key applications: stream analytics, and machine learning training and inference.</div><div><br></div><div>First, we introduce Streambox-TZ (SBT), a secure stream analytics engine for ARM-based edge platforms. SBT contributes a data plane that isolates only analytics’ data and</div><div>computation in a trusted execution environment (TEE). By design, SBT achieves a minimal trusted computing base (TCB) inside TEE, incurring modest security overhead.</div><div><br></div><div>Second, we design a minimal GPU software stack (50KB), called GPURip. GPURip allows developers to record GPU computation ahead of time, which will be replayed later</div><div>on client devices. In doing so, GPURip excludes the original GPU stack from run time eliminating its wide attack surface and exploitable vulnerabilities.</div><div><br></div><div>Finally, we propose CoDry, a novel approach for TEE to record GPU computation remotely. CoDry provides an online GPU recording in a safe and practical way; it hosts GPU stacks in the cloud that collaboratively perform a dryrun with client GPU models. To overcome frequent interactions over a wireless connection, CoDry implements a suite of key optimizations.</div>

The advent of virtualization introduced the need for virtual switches to interconnect virtual machines deployed in a cloud infrastructure. With Software Defined Networking (SDN), a central controller can configure these virtual switches. Virtual switches execute on commodity operating systems. Open vSwitch is an open source project that is widely used in production cloud environments. If an adversary gains access with full privileges to the operating system hosting the virtual switch, then Open vSwitch becomes vulnerable to a variety of different attacks that could compromise the whole network. The purpose of this thesis project is to improve the security of Open vSwitch implementations in order to ensure that only authenticated switches and controllers can communicate with each other, while maintaining code integrity and confidentiality of keys and certificates. The thesis project proposes a design and shows an implementation that leverages Intel® Safe Guard Extensions (SGX) technology. A new library, TLSonSGX, is implemented. This library replaces the use of the OpenSSL library in Open vSwitch. In addition to implementing standard Transport Level Security (TLS) connectivity, TLSonSGX confines TLS communication in the protected memory enclave and hence protects TLS sensitive components necessary to provide confidentiality and integrity, such as private keys and negotiated symmetric keys. Moreover, TLSonSGX introduces new, secure, and automatic means to generate keys and obtain signed certificates from a central Certificate Authority that validates using Linux Integrity Measurements Architecture (IMA) that the Open vSwitch binaries have not been tampered with before issuing a signed certificate. The generated keys and obtained certificates are stored in the memory enclave and hence never exposed as plaintext outside the enclave. This new mechanism is a replacement for the existing manual and unsecure procedures (as described in Open vSwitch project). A security analysis of the system is provided as well as an examination of performance impact of the use of a trusted execution environment. Results show that generating keys and certificates using TLSonSGX takes less than 0.5 seconds while adding 30% latency overhead for the first packet in a flow compared to using OpenSSL when both are executed on Intel® CoreTM i7-6600U processor clocked at 2.6 GHz. These results show that TLSonSGX can enhance Open vSwitch security and reduce its TLS configuration overhead. / Framkomsten av virtualisering införde behovet av virtuella växlar för att koppla tillsammans virtuella maskiner placerade i molninfrastruktur. Med mjukvarubaserad nätverksteknik (SDN), kan ett centralt styrenhet konfigurera dessa virtuella växlar. Virtuella växlar kör på standardoperativsystem. Open vSwitch är ett open-source projekt som ofta används i molntjänster. Om en motståndare får tillgång med fullständiga privilegier till operativsystemet där Open vSwitch körs, blir Open vSwitch utsatt för olika attacker som kan kompromettera hela nätverket.  Syftet med detta examensarbete är att förbättra säkerheten hos Open vSwitch för att garantera att endast autentiserade växlar och styrenheter kan kommunicera med varandra, samtidigt som att upprätthålla kod integritet och konfidentialitet av nycklar och certifikat. Detta examensarbete föreslår en design och visar en implementation som andvändar Intel®s Safe Guard Extensions (SGX) teknologi. Ett nytt bibliotek, TLSonSGX, är implementerat. Detta bibliotek ersätter biblioteket OpenSSL i Open vSwitch. Utöver att det implementerar ett standard “Transport Layer Security” (TLS) anslutning, TLSonSGX begränsar TLS kommunikation i den skyddade minnes enklaven och skyddar därför TLS känsliga komponenter som är nödvändiga för att ge sekretess och integritet, såsom privata nycklar och förhandlade symmetriska nycklar. Dessutom introducerar TLSonSGX nya, säkra och automatiska medel för att generera nycklar och få signerade certifikat från en central certifikatmyndighet som validerar, med hjälp av Linux Integrity Measurements Architecture (IMA), att Open vSwitch-binärerna inte har manipulerats innan de utfärdade ett signerat certifikat. De genererade nycklarna och erhållna certifikat lagras i minnes enklaven och är därför aldrig utsatta utanför enklaven. Denna nya mekanism ersätter de manuella och osäkra procedurerna som beskrivs i Open vSwitch projektet. En säkerhetsanalys av systemet ges såväl som en granskning av prestandaffekten av användningen av en pålitlig exekveringsmiljö. Resultaten visar att använda TLSonSGX för att generera nycklar och certifikat tar mindre än 0,5 sekunder medan det lägger 30% latens overhead för det första paketet i ett flöde jämfört med att använda OpenSSL när båda exekveras på Intel® Core TM processor i7-6600U klockad vid 2,6 GHz. Dessa resultat visar att TLSonSGX kan förbättra Open vSwitch säkerhet och minska TLS konfigurationskostnaden.

We are living in the time of the digital revolution in which the world we know changes beyond recognition every decade. The positive aspect is that these changes also drive the progress in quality and availability of digital assets crucial for our societies. To name a few examples, these are broadly available communication channels allowing quick exchange of knowledge over long distances, systems controlling automatic share and distribution of renewable energy in international power grid networks, easily accessible applications for early disease detection enabling self-examination without burdening the health service, or governmental systems assisting citizens to settle official matters without leaving their homes. Unfortunately, however, digitalization also opens opportunities for malicious actors to threaten our societies if they gain control over these assets after successfully exploiting vulnerabilities in the complex computing systems building them. Protecting these systems, which are called high-assurance security systems, is therefore of utmost importance. For decades, humanity has struggled to find methods to protect high-assurance security systems. The advancements in the computing systems security domain led to the popularization of hardware-assisted security techniques, nowadays available in commodity computers, that opened perspectives for building more sophisticated defense mechanisms at lower costs. However, none of these techniques is a silver bullet. Each one targets particular use cases, suffers from limitations, and is vulnerable to specific attacks. I argue that some of these techniques are synergistic and help overcome limitations and mitigate specific attacks when used together. My reasoning is supported by regulations that legally bind high-assurance security systems' owners to provide strong security guarantees. These requirements can be fulfilled with the help of diverse technologies that have been standardized in the last years. In this thesis, I introduce new techniques for hardening high-assurance security systems that execute in remote execution environments, such as public and hybrid clouds. I implemented these techniques as part of a framework that provides technical assurance that high-assurance security systems execute in a specific data center, on top of a trustworthy operating system, in a virtual machine controlled by a trustworthy hypervisor or in strong isolation from other software. I demonstrated the practicality of my approach by leveraging the framework to harden real-world applications, such as machine learning applications in the eHealth domain. The evaluation shows that the framework is practical. It induces low performance overhead (<6%), supports software updates, requires no changes to the legacy application's source code, and can be tailored to individual trust boundaries with the help of security policies. The framework consists of a decentralized monitoring system that offers better scalability than traditional centralized monitoring systems. Each monitored machine runs a piece of code that verifies that the machine's integrity and geolocation conform to the given security policy. This piece of code, which serves as a trusted anchor on that machine, executes inside the trusted execution environment, i.e., Intel SGX, to protect itself from the untrusted host, and uses trusted computing techniques, such as trusted platform module, secure boot, and integrity measurement architecture, to attest to the load-time and runtime integrity of the surrounding operating system running on a bare metal machine or inside a virtual machine. The trusted anchor implements my novel, formally proven protocol, enabling detection of the TPM cuckoo attack. The framework also implements a key distribution protocol that, depending on the individual security requirements, shares cryptographic keys only with high-assurance security systems executing in the predefined security settings, i.e., inside the trusted execution environments or inside the integrity-enforced operating system. Such an approach is particularly appealing in the context of machine learning systems where some algorithms, like the machine learning model training, require temporal access to large computing power. These algorithms can execute inside a dedicated, trusted data center at higher performance because they are not limited by security features required in the shared execution environment. The evaluation of the framework showed that training of a machine learning model using real-world datasets achieved 0.96x native performance execution on the GPU and a speedup of up to 1560x compared to the state-of-the-art SGX-based system. Finally, I tackled the problem of software updates, which makes the operating system's integrity monitoring unreliable due to false positives, i.e., software updates move the updated system to an unknown (untrusted) state that is reported as an integrity violation. I solved this problem by introducing a proxy to a software repository that sanitizes software packages so that they can be safely installed. The sanitization consists of predicting and certifying the future (after the specific updates are installed) operating system's state. The evaluation of this approach showed that it supports 99.76% of the packages available in Alpine Linux main and community repositories. The framework proposed in this thesis is a step forward in verifying and enforcing that high-assurance security systems execute in an environment compliant with regulations. I anticipate that the framework might be further integrated with industry-standard security information and event management tools as well as other security monitoring mechanisms to provide a comprehensive solution hardening high-assurance security systems.

Federated Learning (FL), one variant of Machine Learning (ML) technology, has emerged as a prevalent method for multiple parties to collaboratively train ML models in a distributed manner with the help of a central server normally supplied by a Cloud Service Provider (CSP). Nevertheless, many existing vulnerabilities pose a threat to the advantages of FL and cause potential risks to data security and privacy, such as data leakage, misuse of the central server, or the threat of eavesdroppers illicitly seeking sensitive information. Promisingly advanced cryptography technologies such as Homomorphic Encryption (HE) and Confidential Computing (CC) can be utilized to enhance the security and privacy of FL. However, the development of a framework that seamlessly combines these technologies together to provide confidential FL while retaining efficiency remains an ongoing challenge. In this degree project, we develop a lightweight and user-friendly FL framework called Heflp, which integrates HE and CC to ensure data confidentiality and integrity throughout the entire FL lifecycle. Heflp supports four HE schemes to fit diverse user requirements, comprising three pre-existing schemes and one optimized scheme that we design, named Flashev2, which achieves the highest time and spatial efficiency across most scenarios. The time and memory overheads of all four HE schemes are also evaluated and a comparison between the pros and cons of each other is summarized. To validate the effectiveness, Heflp is tested on the MNIST dataset and the Threat Intelligence dataset provided by CanaryBit, and the results demonstrate that it successfully preserves data privacy without compromising model accuracy. / Federated Learning (FL), en variant av Maskininlärning (ML)-teknologi, har framträtt som en dominerande metod för flera parter att samarbeta om att distribuerat träna ML-modeller med hjälp av en central server som vanligtvis tillhandahålls av en molntjänstleverantör (CSP). Trots detta utgör många befintliga sårbarheter ett hot mot FL:s fördelar och medför potentiella risker för datasäkerhet och integritet, såsom läckage av data, missbruk av den centrala servern eller risken för avlyssnare som olagligt söker känslig information. Lovande avancerade kryptoteknologier som Homomorf Kryptering (HE) och Konfidentiell Beräkning (CC) kan användas för att förbättra säkerheten och integriteten för FL. Utvecklingen av en ramverk som sömlöst kombinerar dessa teknologier för att erbjuda konfidentiellt FL med bibehållen effektivitet är dock fortfarande en pågående utmaning. I detta examensarbete utvecklar vi en lättviktig och användarvänlig FL-ramverk som kallas Heflp, som integrerar HE och CC för att säkerställa datakonfidentialitet och integritet under hela FLlivscykeln. Heflp stöder fyra HE-scheman för att passa olika användarbehov, bestående av tre befintliga scheman och ett optimerat schema som vi designar, kallat Flashev2, som uppnår högsta tids- och rumeffektivitet i de flesta scenarier. Tids- och minneskostnaderna för alla fyra HE-scheman utvärderas också, och en jämförelse mellan fördelar och nackdelar sammanfattas. För att validera effektiviteten testas Heflp på MNIST-datasetet och Threat Intelligence-datasetet som tillhandahålls av CanaryBit, och resultaten visar att det framgångsrikt bevarar datasekretessen utan att äventyra modellens noggrannhet.

The requirements of privacy have become a necessity in modern-day internet-based applications. This applies from traditional client-server applications to blockchain-based applications. Blockchains being a new domain for application development, the priority towards privacy beyond pseudo anonymity has been lacking. With financial applications built on blockchains entering mainstream adoption, and these applications handling sensitive data of users, it is useful to be able to understand how privacy technologies can help in ensuring that the user’s data privacy is maintained. This project addresses this by taking a simple financial transaction use case and applying various privacy technologies like Data Encryption, Zero-Knowledge Proofs, Trusted Execution Environments. Workflow and Component architecture is proposed for solutions based on these technologies and they are compared to identify which is a feasible solution for the use case. Trusted Execution Environments was concluded to be the best match for the requirements of the use case and Secret Network which is a blockchain built on this privacy technology was evaluated against determined privacy metrics and benchmarks were run to check the performance changes due to using the technology. Based on this analysis, Secret Network was found to be a good solution to handle the provided use case and flexible enough to handle more complex requirements. / Kraven på integritet har blivit en nödvändighet i dagens internetbaserade tillämpningar. Detta gäller från traditionella klient-server-tillämpningar till blockkedjebaserade tillämpningar. Eftersom blockkedjor är ett nytt område för utveckling av tillämpningar har man inte prioriterat integritet utöver pseudoanonymitet. I och med att finansiella tillämpningar som byggs på blockkedjor börjar bli allmänt accepterade, och att dessa tillämpningar hanterar känsliga uppgifter om användarna, är det bra att kunna förstå hur integritetsskyddstekniker kan bidra till att se till att användarnas integritet bevaras. Detta projekt tar itu med detta genom att ta ett enkelt användningsfall för finansiella transaktioner och tillämpa olika integritetsskyddstekniker som datakryptering, bevis för nollkunskap och betrodda utförandemiljöer. Arbetsflöden och komponentarkitektur föreslås för lösningar som bygger på dessa tekniker och de jämförs för att identifiera vilken lösning som är genomförbar för användningsfallet. Trusted Execution Environments konstaterades vara den bästa lösningen för kraven i användningsfallet och Secret Network, som är en blockkedja byggd på denna teknik för skydd av privatlivet, utvärderades mot fastställda integritetsmått och benchmarks kördes för att kontrollera prestandaförändringarna till följd av användningen av tekniken. På grundval av denna analys konstaterades Secret Network vara en bra lösning för att hantera det aktuella användningsfallet och tillräckligt flexibel för att hantera mer komplexa krav.

With recent advances in electronics and communication technologies, our daily lives are immersed in an environment of Internet-connected smart things. Despite the great convenience brought by the development of these technologies, privacy concerns and security issues are two topics that deserve more attention. On one hand, as smart things continue to grow in their abilities to sense the physical world and capabilities to send information out through the Internet, they have the potential to be used for surveillance of any individuals secretly. Nevertheless, people tend to adopt wearable devices without fully understanding what private information can be inferred and leaked through sensor data. On the other hand, security issues become even more serious and lethal with the world embracing the Internet of Things (IoT). Failures in computing systems are common, however, a failure now in IoT may harm people's lives. As demonstrated in both academic research and industrial practice, a software vulnerability hidden in a smart vehicle may lead to a remote attack that subverts a driver's control of the vehicle. Our approach to the aforementioned challenges starts by understanding privacy leakage in the IoT era and follows with adding defense layers to the IoT system with attackers gaining increasing capabilities. The first question we ask ourselves is "what new privacy concerns do IoT bring". We focus on discovering information leakage beyond people's common sense from even seemingly benign signals. We explore how much private information we can extract by designing information extraction systems. Through our research, we argue for stricter access control on newly coming sensors. After noticing the importance of data collected by IoT, we trace where sensitive data goes. In the IoT era, edge nodes are used to process sensitive data. However, a capable attacker may compromise edge nodes. Our second research focuses on applying trusted hardware to build trust in large-scale networks under this circumstance. The application of trusted hardware protects sensitive data from compromised edge nodes. Nonetheless, if an attacker becomes more powerful and embeds malicious logic into code for trusted hardware during the development phase, he still can secretly steal private data. In our third research, we design a static analyzer for detecting malicious logic hidden inside code for trusted hardware. Other than the privacy concern of data collected, another important aspect of IoT is that it affects the physical world. Our last piece of research work enables a user to verify the continuous execution state of an unmanned vehicle. This way, people can trust the integrity of the past and present state of the unmanned vehicle. / Doctor of Philosophy / The past few years have witnessed a rising in computing and networking technologies. Such advances enable the new paradigm, IoT, which brings great convenience to people's life. Large technology companies like Google, Apple, Amazon are creating smart devices such as smartwatch, smart home, drones, etc. Compared to the traditional internet, IoT can provide services beyond digital information by interacting with the physical world by its sensors and actuators. While the deployment of IoT brings value in various aspects of our society, the lucrative reward from cyber-crimes also increases in the upcoming IoT era. Two unique privacy and security concerns are emerging for IoT. On one hand, IoT brings a large volume of new sensors that are deployed ubiquitously and collect data 24/7. User's privacy is a big concern in this circumstance because collected sensor data may be used to infer a user's private activities. On the other hand, cyber-attacks now harm not only cyberspace but also the physical world. A failure in IoT devices could result in loss of human life. For example, a remotely hacked vehicle could shut down its engine on the highway regardless of the driver's operation. Our approach to emerging privacy and security concerns consists of two directions. The first direction targets at privacy protection. We first look at the privacy impact of upcoming ubiquitous sensing and argue for stricter access control on smart devices. Then, we follow the data flow of private data and propose solutions to protect private data from the networking and cloud computing infrastructure. The other direction aims at protecting the physical world. We propose an innovative method to verify the cyber state of IoT devices.